The authors argue mathematically that a common, power-function model of the just-noticeable difference in stimulus intensities is logically inconsistent with an exponent other than 1 in those frequent situations in which a particular averaging over experimental conditions has taken place. The authors show that an alternative power-law model, one which does not share this logical inconsistency, provides a good fit to many well-known, psychoacoustic intensity discrimination data. They also show that the exponent in this alternative model must be nonconstant with the discrimination criterion in experiments implementing this averaging of data.